matched pairs

Questions about the implementation and application of the PLS-SEM method, that are not related to the usage of the SmartPLS software.
Post Reply
drudee
PLS Junior User
Posts: 1
Joined: Thu Jan 14, 2021 9:01 am
Real name and title: Tim Dreesen

matched pairs

Post by drudee »

Hey guys,

probably you can help me.
I’d like to analyse data from two different surveys, the first containing responses from supervisors, the second comprising responses of corresponding team members.

The idea is to measure some constructs with ‘matched pairs’, in particular there are three kind of constructs:
a) Constructs that have been measured by both parties
b) Constructs that have been measured by managers only
c) Constructs which have been measured by team members only

So, a consolidated data table including answers of both surveys would, for example, look like this:
example_datatable.PNG
example_datatable.PNG (42.37 KiB) Viewed 18993 times
No teamID teamPerf leadership congruence_s congruence_tm teamAuton timePress
1 A 4 4 4 na na na
2 A na na na 4 5 5
3 A na na na 4 3 4
4 B 4 2 4 na na na
5 B na na na 4 3 2
6 B na na na 5 3 2
… … … … … … … …

(the columns 'teamPerf' 'leadership' and 'congruence_s' only appear in the survey given to supervisors, 'congruence_tm', 'teamAuton' and 'timePress' belong to the team member survey)

Assuming, I want to measure a possible connection between teamAutonomy (responses from team members only) --> teamPerformance (responses from supervisors only), would the algorithm be able to calculate the regression?

Or do I necessarily have to aggregate data, for example by grouping per teamID?

I was inspired by a research article of Ko, Kirsch and King 2005
(https://www.researchgate.net/publicatio ... mentations) where they have a similar model. But honestly, I have no clue how they prepared (and analysed) the data and what is needed for this kind of analysis.

Do you have any comments or thoughts?
Tim
jmbecker
SmartPLS Developer
Posts: 1282
Joined: Tue Mar 28, 2006 11:09 am
Real name and title: Dr. Jan-Michael Becker

Re: matched pairs

Post by jmbecker »

SmartPLS or PLS algorithm in general need one line (row) of data per observation. If you have a real matched pairs sample then this is possible to analyze with PLS. You have some constructs from the one respondent (e.g., the client) and the other constructs from a matched pair (e.g., the provider). This also seems to be the data structure in the paper you reference. They also seem to have a few constructs with observations from both the client and the provider and seem to average them. I don't know if this is really a good idea, but it got published. I would have modeled this as two separate constructs one for the client and one for the provider.

However, what is not directly possible is to have only one answer from the provider and multiple from the client. It seems that you data with one answer from the team manager and multiple answer from the team members is like this. This is a multi-level or hierarchical data structure. PLS algorithms are not able to handle this appropriately. In this case, application of the PLS method would require averaging the answers from the team members or "copying" the answer from the manager. With that you could get an initial idea about your effects, but it might be better to use some method that can really deal with such multi-level data.
Dr. Jan-Michael Becker, BI Norwegian Business School, SmartPLS Developer
Researchgate: https://www.researchgate.net/profile/Jan_Michael_Becker
GoogleScholar: http://scholar.google.de/citations?user ... AAAJ&hl=de
Post Reply